The meaning and interpretation of ALFF and fALFF

The human brain is never truly silent. Even when we are at rest, eyes closed and mind wandering, its billions of neurons fire in complex, spontaneous patterns. To listen to this quiet hum, scientists use a powerful imaging technique called functional magnetic resonance imaging, or fMRI. This method does not measure the electrical sparks of neurons directly; instead, it tracks the subtle changes in blood flow and oxygen levels that follow neural activity. When a group of brain cells becomes active, they demand more oxygen, and the body rushes blood to that area. By measuring these tiny fluctuations in blood oxygen, researchers can create a map of the brain's resting state, revealing which regions are talking to one another.

For years, scientists have relied on two specific numbers to describe the intensity of these resting brain signals. One is a measure of the raw strength of the low-frequency wiggles in the blood signal, while the other is a normalized version of that same number, designed to cancel out background noise. These metrics have become standard tools for comparing healthy brains to those affected by disease, or for studying how the brain changes with age. However, a fundamental question has lingered: do these numbers actually tell us about the brain's neural activity, or are they merely reflecting the quirks of the machine taking the picture?

A team of researchers at Vanderbilt University has now answered this question by building a detailed mathematical model that connects the dots between brain metabolism and the signals seen on the scanner. They discovered that the two common metrics, often treated as interchangeable, actually measure very different things. The raw measure of signal strength is closely tied to the brain's baseline metabolism, scaling predictably with how much oxygen the brain uses. In contrast, the normalized version, which was thought to be a cleaner, more reliable indicator, is actually trapped in a narrow range determined almost entirely by the settings of the scanner itself.

The researchers began by revising a well-known model of how blood flows through the brain's vessels, known as the "balloon model." They derived a precise formula that describes how changes in the brain's oxygen consumption translate into the blood-oxygen signals detected by the MRI machine. Using this formula, they simulated the resting-state signals that would appear in a typical scan, accounting for the natural fluctuations in brain metabolism and the inevitable thermal noise that every electronic device produces. They then calculated what the two standard metrics would look like under these conditions, treating the brain's activity as a continuous, low-frequency hum.

Their analysis revealed a striking limitation in the normalized metric. They found that this value is mathematically confined between a lower floor and an upper ceiling. The floor is set by the fraction of the total frequency range that the researcher chooses to analyze, while the ceiling is determined by how much of the true brain signal falls within that specific range. Crucially, both of these boundaries are fixed by the acquisition parameters: how fast the scanner takes pictures and how long the scan lasts. The researchers showed that if you change the speed at which the scanner captures images, the value of this metric shifts dramatically, even if the brain itself remains exactly the same.

In their simulations, they tested what happens when the time between images is shortened from two seconds to half a second. With the brain's activity held constant, the normalized value dropped by nearly forty-five percent. This means that a researcher comparing data from two different studies, or even two different scanners, could see a massive difference in this number that has nothing to do with the patient's brain function and everything to do with the machine's settings. The value is so tightly bound to these technical choices that it cannot be used to compare results across different protocols.

The raw signal strength, on the other hand, behaves differently. While it is also affected by noise, it is not trapped by the same rigid boundaries. The researchers demonstrated that this measure follows a predictable power law relationship with the brain's baseline metabolism. If the brain's oxygen consumption increases, the raw signal strength increases in a consistent, calculable way. This makes it a far more sensitive and reliable indicator of the brain's underlying metabolic state, provided the signal is strong enough to rise above the noise.

The study also highlighted how the duration of the scan influences the results. As a scan gets longer, the maximum possible value for the normalized metric actually decreases. This is because a longer scan captures more of the very low-frequency power of the brain's natural rhythm, which sits outside the specific band researchers usually analyze. This adds more "total" power to the denominator of the calculation, effectively lowering the final number. This effect is systematic and predictable, yet it contradicts the common intuition that longer scans always yield better, higher-quality estimates.

The authors conclude that the normalized metric is not a direct reflection of neural activity intensity, but rather a bounded readout of the signal-to-noise ratio. At the typical signal strengths found in human brain scans, this value sits very close to its lower limit, meaning that most of the variation seen in maps of this metric likely reflects differences in noise or scanner settings rather than differences in brain function. The researchers argue that treating these two metrics as interchangeable alternatives is a mistake. They suggest that if scientists wish to use the normalized metric, they must report the specific scanner settings and the calculated boundaries alongside their results, so that readers understand the context of the numbers.

Ultimately, this work clarifies what these popular brain maps are actually showing. The raw measure of signal amplitude offers a window into the brain's metabolic engine, scaling with how hard the brain is working at rest. The normalized measure, while useful for reducing certain types of noise, is heavily constrained by the physics of the measurement itself. It cannot be used to compare brains across different studies unless the scanning protocols are identical. By exposing these hidden dependencies, the researchers provide a clearer path for interpreting the silent, rhythmic language of the resting brain.